FW: Differential incremental backups

From: Mark W. Farnham <mwf_at_rsiz.com>
Date: Thu, 18 Mar 2010 14:20:41 -0400
Message-ID: <B57AE9FC96974CCFA6ADE5CE579D3272_at_rsiz.com>

snipped to fit  

From: Mark W. Farnham [mailto:mwf_at_rsiz.com] Sent: Thursday, March 18, 2010 1:49 PM
To: 'oralrnr_at_gmail.com'; 'Guillermo Alan Bort' Cc: 'andrey.hudyakov_at_gmail.com'; 'william.muriithi_at_epicadvertising.com'; 'oracle-l_at_freelists.org'
Subject: RE: Differential incremental backups  

Are you saying you no longer have the previous full backup?  

All I can think of then is if you have flashback database enabled and the Friday before the Sunday base backup is still in range you would flash back. You wouldn't restore at all.  

IF you have a full backup prior to the point in time you need to recover to you restore the full backup that is closest in time to, but BEFORE the point in time to which you want to recover. The incrementals are indeed with respect to some full backup. Then you apply the cheapest incrementals (if any, and see Robert's note to understand what I mean by cheapest) still prior to the time to which you want to recover, and then the archived and/or online redo logs take it from there to the resetlogs point in time you need. (Which is why you should preserve a copy of the online redo logs as the first step of the recovery process so you can put them back in place and try again if you or the hardware screws up at an awkward moment during recovery. This being a separate notion entirely from the religious war about whether or not to back up online redo logs routinely which jihad I will not engage in.) As of my last test, RMAN does NOT preserve the online redo logs for you in this scenario. It probably should. All the other stuff RMAN is supposed to figure out for you correctly without much thinking.  

Are you in an actual recovery, or are you just trying to figure out if you can whack something by hand that RMAN doesn't want to let go of?  

Regarding that: You're in way over your head. I wouldn't whack anything RMAN says you should keep unless it corresponds EXACTLY to a documented bug that says you are supposed to.  

So if

Sunday March 7 full backup

<several days of incrementals call it set A for reference>

Sunday March 14 full backup

<several days of incrementals call it set B for reference>

and now you want to go back to Friday, March 12, then you better have the March 7 full backup available to make the set A incrementals relevant.  

Robert described the types of incrementals nicely, and Alan mentioned that the recovery window settings would keep the older full backup around (at least through RMAN, meaning you don't intervene with sufficient permissions and zap things from the OS level.)  

So if your window is seven days and today is Thursday, March 18, then RMAN will have kept the full backup from March 7, even though that is more than 7 days ago.  

So in a sense you are right: Incrementals are with respect to a full that was made prior to the incremental.  

All yall please correct me if I botched something. I typed this pretty fast.  



From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Orlando L
Sent: Thursday, March 18, 2010 11:21 AM
To: Guillermo Alan Bort
Cc: andrey.hudyakov_at_gmail.com; william.muriithi_at_epicadvertising.com; oracle-l_at_freelists.org
Subject: Re: Differential incremental backups    

Thanks. I wonder how it can restore to previous Friday if a base backup gets taken Sunday night. I thought it would restore base backup first and then apply incremental on the top of that base backup. I thought the backups taken before base backup were useless and cannot be used to restore to a point in time in last week, even if recovery window of 7 days is specified. ie, in this example, wanting to go back to last Friday on a given Monday, with a base backup on Sunday.  

Received on Thu Mar 18 2010 - 13:20:41 CDT

Original text of this message